Skip to content

[fe bug] query information_schema.tables cost 7s, even more  #4824

Closed

Description

@francisoliverlee

Describe the bug
exec the sql cost 7s , even more

select * from information_schema.tables

To Reproduce
exec the sql

select * from information_schema.tables

Expected behavior
cost less than 1s

Activity

  1. francisoliverlee commented on Oct 30, 2020

    @francisoliverlee
    MemberAuthor

    after some check, it's known to us how this happened. how to exec the sql between fe and be

    we user  --------> fe --------------> be --------------------------------------------------------> fe
              sql           sql plan                  rpc call to get information_schema data from fe
    

    when socket between FE and BE for rpc call timeout and BE calling fe, FE would send TCP RST command to BE, BE would retry ( after 1s in default, can be configed for thrift_client_retry_interval_ms ).

    the retry is the COST.

  2. caiconghui commented on Oct 30, 2020

    @caiconghui
    Contributor

    @francisoliverlee but I am stiil confused that even thrift client retry why cost so much time about 7s?

  3. francisoliverlee commented on Nov 1, 2020

    @francisoliverlee
    MemberAuthor

    @francisoliverlee but I am stiil confused that even thrift client retry why cost so much time about 7s?

    so do i. until now i can point out it is the RPC-CALL-FROM-BE-2-FE cost more than 1s. there are 2 prof

    prof 1:

    `---[2031.87977ms] org.apache.doris.qe.ResultReceiver:getNext()
            +---[0.001568ms] org.apache.doris.qe.RowBatch:<init>() #65
            +---[0.001786ms] org.apache.doris.rpc.PFetchDataRequest:<init>() #68
            +---[0.002026ms] org.apache.doris.rpc.BackendServiceProxy:getInstance() #71
            +---[0.018739ms] org.apache.doris.rpc.BackendServiceProxy:fetchDataAsync() #71
            +---[0.003719ms] org.apache.doris.thrift.TStatusCode:findByValue() #89
            +---[0.00269ms] org.apache.doris.qe.RowBatch:setQueryStatistics() #95
            +---[0.00213ms] org.apache.doris.rpc.PFetchDataRequest:getSerializedResult() #106
            +---[0.001838ms] org.apache.doris.thrift.TResultBatch:<init>() #108
            +---[0.001712ms] org.apache.thrift.TDeserializer:<init>() #109
            +---[0.003755ms] org.apache.thrift.TDeserializer:deserialize() #110
            +---[0.001832ms] org.apache.doris.qe.RowBatch:setBatch() #111
            `---[0.00172ms] org.apache.doris.qe.RowBatch:setEos() #112
    

    prof 2

    rpc api exec in FE like getDbNames(), listTableStatus().... cost less than 1ms.(i did the cost-check, but lost the content)
    

    there are 2 possibilities

    1. BE has some lock or something delay to call FE
    2. FE delay to exec RPC-call
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions